Amharic Character Recognition using a Fast Signature Based Algorithm

The Amharic language is the principal language of over 20 million people mainly in Ethiopia. An extensive literature survey reveals no journal or conference papers on Amharic character recognition. The Amharic script has 33 basic characters each with seven orders giving 231 distinct characters, not including numbers and punctuation symbols. The characters are cursive but not connected and unlike other cursive scripts do not use dots. This paper describes the Amharic script and discusses the difficulties of applying conventional structural and syntactic recognition processes. Two statistical algorithms for identifying Amharic characters are described. In both, the characters are normalised for both size and orientation. The first compares the character against a series of templates. The second derives a characteristic signature from the character and compares this against a set of signature templates. The signatures used are fifty times smaller than the original character and the recognition process is corresponding faster but with some loss of accuracy. The statistical techniques described have been fully implemented and the resulting performance outlined.